Linux Audio

Check our new training course

Loading...
v4.17
  1/*
  2 *  linux/arch/arm/mm/proc-xscale.S
  3 *
  4 *  Author:	Nicolas Pitre
  5 *  Created:	November 2000
  6 *  Copyright:	(C) 2000, 2001 MontaVista Software Inc.
  7 *
  8 * This program is free software; you can redistribute it and/or modify
  9 * it under the terms of the GNU General Public License version 2 as
 10 * published by the Free Software Foundation.
 11 *
 12 * MMU functions for the Intel XScale CPUs
 13 *
 14 * 2001 Aug 21:
 15 *	some contributions by Brett Gaines <brett.w.gaines@intel.com>
 16 *	Copyright 2001 by Intel Corp.
 17 *
 18 * 2001 Sep 08:
 19 *	Completely revisited, many important fixes
 20 *	Nicolas Pitre <nico@fluxnic.net>
 21 */
 22
 23#include <linux/linkage.h>
 24#include <linux/init.h>
 25#include <asm/assembler.h>
 26#include <asm/hwcap.h>
 27#include <asm/pgtable.h>
 28#include <asm/pgtable-hwdef.h>
 29#include <asm/page.h>
 30#include <asm/ptrace.h>
 31#include "proc-macros.S"
 32
 33/*
 34 * This is the maximum size of an area which will be flushed.  If the area
 35 * is larger than this, then we flush the whole cache
 36 */
 37#define MAX_AREA_SIZE	32768
 38
 39/*
 40 * the cache line size of the I and D cache
 41 */
 42#define CACHELINESIZE	32
 43
 44/*
 45 * the size of the data cache
 46 */
 47#define CACHESIZE	32768
 48
 49/*
 50 * Virtual address used to allocate the cache when flushed
 51 *
 52 * This must be an address range which is _never_ used.  It should
 53 * apparently have a mapping in the corresponding page table for
 54 * compatibility with future CPUs that _could_ require it.  For instance we
 55 * don't care.
 56 *
 57 * This must be aligned on a 2*CACHESIZE boundary.  The code selects one of
 58 * the 2 areas in alternance each time the clean_d_cache macro is used.
 59 * Without this the XScale core exhibits cache eviction problems and no one
 60 * knows why.
 61 *
 62 * Reminder: the vector table is located at 0xffff0000-0xffff0fff.
 63 */
 64#define CLEAN_ADDR	0xfffe0000
 65
 66/*
 67 * This macro is used to wait for a CP15 write and is needed
 68 * when we have to ensure that the last operation to the co-pro
 69 * was completed before continuing with operation.
 70 */
 71	.macro	cpwait, rd
 72	mrc	p15, 0, \rd, c2, c0, 0		@ arbitrary read of cp15
 73	mov	\rd, \rd			@ wait for completion
 74	sub 	pc, pc, #4			@ flush instruction pipeline
 75	.endm
 76
 77	.macro	cpwait_ret, lr, rd
 78	mrc	p15, 0, \rd, c2, c0, 0		@ arbitrary read of cp15
 79	sub	pc, \lr, \rd, LSR #32		@ wait for completion and
 80						@ flush instruction pipeline
 81	.endm
 82
 83/*
 84 * This macro cleans the entire dcache using line allocate.
 85 * The main loop has been unrolled to reduce loop overhead.
 86 * rd and rs are two scratch registers.
 87 */
 88	.macro  clean_d_cache, rd, rs
 89	ldr	\rs, =clean_addr
 90	ldr	\rd, [\rs]
 91	eor	\rd, \rd, #CACHESIZE
 92	str	\rd, [\rs]
 93	add	\rs, \rd, #CACHESIZE
 941:	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
 95	add	\rd, \rd, #CACHELINESIZE
 96	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
 97	add	\rd, \rd, #CACHELINESIZE
 98	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
 99	add	\rd, \rd, #CACHELINESIZE
100	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
101	add	\rd, \rd, #CACHELINESIZE
102	teq	\rd, \rs
103	bne	1b
104	.endm
105
106	.data
107	.align	2
108clean_addr:	.word	CLEAN_ADDR
109
110	.text
111
112/*
113 * cpu_xscale_proc_init()
114 *
115 * Nothing too exciting at the moment
116 */
117ENTRY(cpu_xscale_proc_init)
118	@ enable write buffer coalescing. Some bootloader disable it
119	mrc	p15, 0, r1, c1, c0, 1
120	bic	r1, r1, #1
121	mcr	p15, 0, r1, c1, c0, 1
122	ret	lr
123
124/*
125 * cpu_xscale_proc_fin()
126 */
127ENTRY(cpu_xscale_proc_fin)
128	mrc	p15, 0, r0, c1, c0, 0		@ ctrl register
129	bic	r0, r0, #0x1800			@ ...IZ...........
130	bic	r0, r0, #0x0006			@ .............CA.
131	mcr	p15, 0, r0, c1, c0, 0		@ disable caches
132	ret	lr
133
134/*
135 * cpu_xscale_reset(loc)
136 *
137 * Perform a soft reset of the system.  Put the CPU into the
138 * same state as it would be if it had been reset, and branch
139 * to what would be the reset vector.
140 *
141 * loc: location to jump to for soft reset
142 *
143 * Beware PXA270 erratum E7.
144 */
145	.align	5
146	.pushsection	.idmap.text, "ax"
147ENTRY(cpu_xscale_reset)
148	mov	r1, #PSR_F_BIT|PSR_I_BIT|SVC_MODE
149	msr	cpsr_c, r1			@ reset CPSR
150	mcr	p15, 0, r1, c10, c4, 1		@ unlock I-TLB
151	mcr	p15, 0, r1, c8, c5, 0		@ invalidate I-TLB
152	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
153	bic	r1, r1, #0x0086			@ ........B....CA.
154	bic	r1, r1, #0x3900			@ ..VIZ..S........
155	sub	pc, pc, #4			@ flush pipeline
156	@ *** cache line aligned ***
157	mcr	p15, 0, r1, c1, c0, 0		@ ctrl register
158	bic	r1, r1, #0x0001			@ ...............M
159	mcr	p15, 0, ip, c7, c7, 0		@ invalidate I,D caches & BTB
160	mcr	p15, 0, r1, c1, c0, 0		@ ctrl register
161	@ CAUTION: MMU turned off from this point. We count on the pipeline
162	@ already containing those two last instructions to survive.
163	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
164	ret	r0
165ENDPROC(cpu_xscale_reset)
166	.popsection
167
168/*
169 * cpu_xscale_do_idle()
170 *
171 * Cause the processor to idle
172 *
173 * For now we do nothing but go to idle mode for every case
174 *
175 * XScale supports clock switching, but using idle mode support
176 * allows external hardware to react to system state changes.
177 */
178	.align	5
179
180ENTRY(cpu_xscale_do_idle)
181	mov	r0, #1
182	mcr	p14, 0, r0, c7, c0, 0		@ Go to IDLE
183	ret	lr
184
185/* ================================= CACHE ================================ */
186
187/*
188 *	flush_icache_all()
189 *
190 *	Unconditionally clean and invalidate the entire icache.
191 */
192ENTRY(xscale_flush_icache_all)
193	mov	r0, #0
194	mcr	p15, 0, r0, c7, c5, 0		@ invalidate I cache
195	ret	lr
196ENDPROC(xscale_flush_icache_all)
197
198/*
199 *	flush_user_cache_all()
200 *
201 *	Invalidate all cache entries in a particular address
202 *	space.
203 */
204ENTRY(xscale_flush_user_cache_all)
205	/* FALLTHROUGH */
206
207/*
208 *	flush_kern_cache_all()
209 *
210 *	Clean and invalidate the entire cache.
211 */
212ENTRY(xscale_flush_kern_cache_all)
213	mov	r2, #VM_EXEC
214	mov	ip, #0
215__flush_whole_cache:
216	clean_d_cache r0, r1
217	tst	r2, #VM_EXEC
218	mcrne	p15, 0, ip, c7, c5, 0		@ Invalidate I cache & BTB
219	mcrne	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
220	ret	lr
221
222/*
223 *	flush_user_cache_range(start, end, vm_flags)
224 *
225 *	Invalidate a range of cache entries in the specified
226 *	address space.
227 *
228 *	- start - start address (may not be aligned)
229 *	- end	- end address (exclusive, may not be aligned)
230 *	- vma	- vma_area_struct describing address space
231 */
232	.align	5
233ENTRY(xscale_flush_user_cache_range)
234	mov	ip, #0
235	sub	r3, r1, r0			@ calculate total size
236	cmp	r3, #MAX_AREA_SIZE
237	bhs	__flush_whole_cache
238
2391:	tst	r2, #VM_EXEC
240	mcrne	p15, 0, r0, c7, c5, 1		@ Invalidate I cache line
241	mcr	p15, 0, r0, c7, c10, 1		@ Clean D cache line
242	mcr	p15, 0, r0, c7, c6, 1		@ Invalidate D cache line
243	add	r0, r0, #CACHELINESIZE
244	cmp	r0, r1
245	blo	1b
246	tst	r2, #VM_EXEC
247	mcrne	p15, 0, ip, c7, c5, 6		@ Invalidate BTB
248	mcrne	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
249	ret	lr
250
251/*
252 *	coherent_kern_range(start, end)
253 *
254 *	Ensure coherency between the Icache and the Dcache in the
255 *	region described by start.  If you have non-snooping
256 *	Harvard caches, you need to implement this function.
257 *
258 *	- start  - virtual start address
259 *	- end	 - virtual end address
260 *
261 *	Note: single I-cache line invalidation isn't used here since
262 *	it also trashes the mini I-cache used by JTAG debuggers.
263 */
264ENTRY(xscale_coherent_kern_range)
265	bic	r0, r0, #CACHELINESIZE - 1
2661:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
267	add	r0, r0, #CACHELINESIZE
268	cmp	r0, r1
269	blo	1b
270	mov	r0, #0
271	mcr	p15, 0, r0, c7, c5, 0		@ Invalidate I cache & BTB
272	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
273	ret	lr
274
275/*
276 *	coherent_user_range(start, end)
277 *
278 *	Ensure coherency between the Icache and the Dcache in the
279 *	region described by start.  If you have non-snooping
280 *	Harvard caches, you need to implement this function.
281 *
282 *	- start  - virtual start address
283 *	- end	 - virtual end address
284 */
285ENTRY(xscale_coherent_user_range)
286	bic	r0, r0, #CACHELINESIZE - 1
2871:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
288	mcr	p15, 0, r0, c7, c5, 1		@ Invalidate I cache entry
289	add	r0, r0, #CACHELINESIZE
290	cmp	r0, r1
291	blo	1b
292	mov	r0, #0
293	mcr	p15, 0, r0, c7, c5, 6		@ Invalidate BTB
294	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
295	ret	lr
296
297/*
298 *	flush_kern_dcache_area(void *addr, size_t size)
299 *
300 *	Ensure no D cache aliasing occurs, either with itself or
301 *	the I cache
302 *
303 *	- addr	- kernel address
304 *	- size	- region size
305 */
306ENTRY(xscale_flush_kern_dcache_area)
307	add	r1, r0, r1
3081:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
309	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
310	add	r0, r0, #CACHELINESIZE
311	cmp	r0, r1
312	blo	1b
313	mov	r0, #0
314	mcr	p15, 0, r0, c7, c5, 0		@ Invalidate I cache & BTB
315	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
316	ret	lr
317
318/*
319 *	dma_inv_range(start, end)
320 *
321 *	Invalidate (discard) the specified virtual address range.
322 *	May not write back any entries.  If 'start' or 'end'
323 *	are not cache line aligned, those lines must be written
324 *	back.
325 *
326 *	- start  - virtual start address
327 *	- end	 - virtual end address
328 */
329xscale_dma_inv_range:
330	tst	r0, #CACHELINESIZE - 1
331	bic	r0, r0, #CACHELINESIZE - 1
332	mcrne	p15, 0, r0, c7, c10, 1		@ clean D entry
333	tst	r1, #CACHELINESIZE - 1
334	mcrne	p15, 0, r1, c7, c10, 1		@ clean D entry
3351:	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
336	add	r0, r0, #CACHELINESIZE
337	cmp	r0, r1
338	blo	1b
339	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
340	ret	lr
341
342/*
343 *	dma_clean_range(start, end)
344 *
345 *	Clean the specified virtual address range.
346 *
347 *	- start  - virtual start address
348 *	- end	 - virtual end address
349 */
350xscale_dma_clean_range:
351	bic	r0, r0, #CACHELINESIZE - 1
3521:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
353	add	r0, r0, #CACHELINESIZE
354	cmp	r0, r1
355	blo	1b
356	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
357	ret	lr
358
359/*
360 *	dma_flush_range(start, end)
361 *
362 *	Clean and invalidate the specified virtual address range.
363 *
364 *	- start  - virtual start address
365 *	- end	 - virtual end address
366 */
367ENTRY(xscale_dma_flush_range)
368	bic	r0, r0, #CACHELINESIZE - 1
3691:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
370	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
371	add	r0, r0, #CACHELINESIZE
372	cmp	r0, r1
373	blo	1b
374	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
375	ret	lr
376
377/*
378 *	dma_map_area(start, size, dir)
379 *	- start	- kernel virtual start address
380 *	- size	- size of region
381 *	- dir	- DMA direction
382 */
383ENTRY(xscale_dma_map_area)
384	add	r1, r1, r0
385	cmp	r2, #DMA_TO_DEVICE
386	beq	xscale_dma_clean_range
387	bcs	xscale_dma_inv_range
388	b	xscale_dma_flush_range
389ENDPROC(xscale_dma_map_area)
390
391/*
392 *	dma_map_area(start, size, dir)
393 *	- start	- kernel virtual start address
394 *	- size	- size of region
395 *	- dir	- DMA direction
396 */
397ENTRY(xscale_80200_A0_A1_dma_map_area)
398	add	r1, r1, r0
399	teq	r2, #DMA_TO_DEVICE
400	beq	xscale_dma_clean_range
401	b	xscale_dma_flush_range
402ENDPROC(xscale_80200_A0_A1_dma_map_area)
403
404/*
405 *	dma_unmap_area(start, size, dir)
406 *	- start	- kernel virtual start address
407 *	- size	- size of region
408 *	- dir	- DMA direction
409 */
410ENTRY(xscale_dma_unmap_area)
411	ret	lr
412ENDPROC(xscale_dma_unmap_area)
413
414	.globl	xscale_flush_kern_cache_louis
415	.equ	xscale_flush_kern_cache_louis, xscale_flush_kern_cache_all
416
417	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
418	define_cache_functions xscale
419
420/*
421 * On stepping A0/A1 of the 80200, invalidating D-cache by line doesn't
422 * clear the dirty bits, which means that if we invalidate a dirty line,
423 * the dirty data can still be written back to external memory later on.
424 *
425 * The recommended workaround is to always do a clean D-cache line before
426 * doing an invalidate D-cache line, so on the affected processors,
427 * dma_inv_range() is implemented as dma_flush_range().
428 *
429 * See erratum #25 of "Intel 80200 Processor Specification Update",
430 * revision January 22, 2003, available at:
431 *     http://www.intel.com/design/iio/specupdt/273415.htm
432 */
433.macro a0_alias basename
434	.globl xscale_80200_A0_A1_\basename
435	.type xscale_80200_A0_A1_\basename , %function
436	.equ xscale_80200_A0_A1_\basename , xscale_\basename
437.endm
438
439/*
440 * Most of the cache functions are unchanged for these processor revisions.
441 * Export suitable alias symbols for the unchanged functions:
442 */
443	a0_alias flush_icache_all
444	a0_alias flush_user_cache_all
445	a0_alias flush_kern_cache_all
446	a0_alias flush_kern_cache_louis
447	a0_alias flush_user_cache_range
448	a0_alias coherent_kern_range
449	a0_alias coherent_user_range
450	a0_alias flush_kern_dcache_area
451	a0_alias dma_flush_range
452	a0_alias dma_unmap_area
453
454	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
455	define_cache_functions xscale_80200_A0_A1
456
457ENTRY(cpu_xscale_dcache_clean_area)
4581:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
459	add	r0, r0, #CACHELINESIZE
460	subs	r1, r1, #CACHELINESIZE
461	bhi	1b
462	ret	lr
463
464/* =============================== PageTable ============================== */
465
466/*
467 * cpu_xscale_switch_mm(pgd)
468 *
469 * Set the translation base pointer to be as described by pgd.
470 *
471 * pgd: new page tables
472 */
473	.align	5
474ENTRY(cpu_xscale_switch_mm)
475	clean_d_cache r1, r2
476	mcr	p15, 0, ip, c7, c5, 0		@ Invalidate I cache & BTB
477	mcr	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
478	mcr	p15, 0, r0, c2, c0, 0		@ load page table pointer
479	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
480	cpwait_ret lr, ip
481
482/*
483 * cpu_xscale_set_pte_ext(ptep, pte, ext)
484 *
485 * Set a PTE and flush it out
486 *
487 * Errata 40: must set memory to write-through for user read-only pages.
488 */
489cpu_xscale_mt_table:
490	.long	0x00						@ L_PTE_MT_UNCACHED
491	.long	PTE_BUFFERABLE					@ L_PTE_MT_BUFFERABLE
492	.long	PTE_CACHEABLE					@ L_PTE_MT_WRITETHROUGH
493	.long	PTE_CACHEABLE | PTE_BUFFERABLE			@ L_PTE_MT_WRITEBACK
494	.long	PTE_EXT_TEX(1) | PTE_BUFFERABLE			@ L_PTE_MT_DEV_SHARED
495	.long	0x00						@ unused
496	.long	PTE_EXT_TEX(1) | PTE_CACHEABLE			@ L_PTE_MT_MINICACHE
497	.long	PTE_EXT_TEX(1) | PTE_CACHEABLE | PTE_BUFFERABLE	@ L_PTE_MT_WRITEALLOC
498	.long	0x00						@ unused
499	.long	PTE_BUFFERABLE					@ L_PTE_MT_DEV_WC
500	.long	0x00						@ unused
501	.long	PTE_CACHEABLE | PTE_BUFFERABLE			@ L_PTE_MT_DEV_CACHED
502	.long	0x00						@ L_PTE_MT_DEV_NONSHARED
503	.long	0x00						@ unused
504	.long	0x00						@ unused
505	.long	0x00						@ unused
506
507	.align	5
508ENTRY(cpu_xscale_set_pte_ext)
509	xscale_set_pte_ext_prologue
510
511	@
512	@ Erratum 40: must set memory to write-through for user read-only pages
513	@
514	and	ip, r1, #(L_PTE_MT_MASK | L_PTE_USER | L_PTE_RDONLY) & ~(4 << 2)
515	teq	ip, #L_PTE_MT_WRITEBACK | L_PTE_USER | L_PTE_RDONLY
516
517	moveq	r1, #L_PTE_MT_WRITETHROUGH
518	and	r1, r1, #L_PTE_MT_MASK
519	adr	ip, cpu_xscale_mt_table
520	ldr	ip, [ip, r1]
521	bic	r2, r2, #0x0c
522	orr	r2, r2, ip
523
524	xscale_set_pte_ext_epilogue
525	ret	lr
526
527	.ltorg
528	.align
529
530.globl	cpu_xscale_suspend_size
531.equ	cpu_xscale_suspend_size, 4 * 6
532#ifdef CONFIG_ARM_CPU_SUSPEND
533ENTRY(cpu_xscale_do_suspend)
534	stmfd	sp!, {r4 - r9, lr}
535	mrc	p14, 0, r4, c6, c0, 0	@ clock configuration, for turbo mode
536	mrc	p15, 0, r5, c15, c1, 0	@ CP access reg
537	mrc	p15, 0, r6, c13, c0, 0	@ PID
538	mrc	p15, 0, r7, c3, c0, 0	@ domain ID
539	mrc	p15, 0, r8, c1, c0, 1	@ auxiliary control reg
540	mrc	p15, 0, r9, c1, c0, 0	@ control reg
541	bic	r4, r4, #2		@ clear frequency change bit
542	stmia	r0, {r4 - r9}		@ store cp regs
543	ldmfd	sp!, {r4 - r9, pc}
544ENDPROC(cpu_xscale_do_suspend)
545
546ENTRY(cpu_xscale_do_resume)
547	ldmia	r0, {r4 - r9}		@ load cp regs
548	mov	ip, #0
549	mcr	p15, 0, ip, c8, c7, 0	@ invalidate I & D TLBs
550	mcr	p15, 0, ip, c7, c7, 0	@ invalidate I & D caches, BTB
551	mcr	p14, 0, r4, c6, c0, 0	@ clock configuration, turbo mode.
552	mcr	p15, 0, r5, c15, c1, 0	@ CP access reg
553	mcr	p15, 0, r6, c13, c0, 0	@ PID
554	mcr	p15, 0, r7, c3, c0, 0	@ domain ID
555	mcr	p15, 0, r1, c2, c0, 0	@ translation table base addr
556	mcr	p15, 0, r8, c1, c0, 1	@ auxiliary control reg
557	mov	r0, r9			@ control register
558	b	cpu_resume_mmu
559ENDPROC(cpu_xscale_do_resume)
560#endif
561
562	.type	__xscale_setup, #function
563__xscale_setup:
564	mcr	p15, 0, ip, c7, c7, 0		@ invalidate I, D caches & BTB
565	mcr	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
566	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I, D TLBs
567	mov	r0, #1 << 6			@ cp6 for IOP3xx and Bulverde
568	orr	r0, r0, #1 << 13		@ Its undefined whether this
569	mcr	p15, 0, r0, c15, c1, 0		@ affects USR or SVC modes
570
571	adr	r5, xscale_crval
572	ldmia	r5, {r5, r6}
573	mrc	p15, 0, r0, c1, c0, 0		@ get control register
574	bic	r0, r0, r5
575	orr	r0, r0, r6
576	ret	lr
577	.size	__xscale_setup, . - __xscale_setup
578
579	/*
580	 *  R
581	 * .RVI ZFRS BLDP WCAM
582	 * ..11 1.01 .... .101
583	 * 
584	 */
585	.type	xscale_crval, #object
586xscale_crval:
587	crval	clear=0x00003b07, mmuset=0x00003905, ucset=0x00001900
588
589	__INITDATA
590
591	@ define struct processor (see <asm/proc-fns.h> and proc-macros.S)
592	define_processor_functions xscale, dabort=v5t_early_abort, pabort=legacy_pabort, suspend=1
593
594	.section ".rodata"
595
596	string	cpu_arch_name, "armv5te"
597	string	cpu_elf_name, "v5"
598
599	string	cpu_80200_A0_A1_name, "XScale-80200 A0/A1"
600	string	cpu_80200_name, "XScale-80200"
601	string	cpu_80219_name, "XScale-80219"
602	string	cpu_8032x_name, "XScale-IOP8032x Family"
603	string	cpu_8033x_name, "XScale-IOP8033x Family"
604	string	cpu_pxa250_name, "XScale-PXA250"
605	string	cpu_pxa210_name, "XScale-PXA210"
606	string	cpu_ixp42x_name, "XScale-IXP42x Family"
607	string	cpu_ixp43x_name, "XScale-IXP43x Family"
608	string	cpu_ixp46x_name, "XScale-IXP46x Family"
609	string	cpu_ixp2400_name, "XScale-IXP2400"
610	string	cpu_ixp2800_name, "XScale-IXP2800"
611	string	cpu_pxa255_name, "XScale-PXA255"
612	string	cpu_pxa270_name, "XScale-PXA270"
613
614	.align
615
616	.section ".proc.info.init", #alloc
617
618.macro xscale_proc_info name:req, cpu_val:req, cpu_mask:req, cpu_name:req, cache
619	.type	__\name\()_proc_info,#object
620__\name\()_proc_info:
621	.long	\cpu_val
622	.long	\cpu_mask
623	.long	PMD_TYPE_SECT | \
624		PMD_SECT_BUFFERABLE | \
625		PMD_SECT_CACHEABLE | \
626		PMD_SECT_AP_WRITE | \
627		PMD_SECT_AP_READ
628	.long	PMD_TYPE_SECT | \
629		PMD_SECT_AP_WRITE | \
630		PMD_SECT_AP_READ
631	initfn	__xscale_setup, __\name\()_proc_info
632	.long	cpu_arch_name
633	.long	cpu_elf_name
634	.long	HWCAP_SWP|HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT|HWCAP_EDSP
635	.long	\cpu_name
636	.long	xscale_processor_functions
637	.long	v4wbi_tlb_fns
638	.long	xscale_mc_user_fns
639	.ifb \cache
640		.long	xscale_cache_fns
641	.else
642		.long	\cache
643	.endif
644	.size	__\name\()_proc_info, . - __\name\()_proc_info
645.endm
646
647	xscale_proc_info 80200_A0_A1, 0x69052000, 0xfffffffe, cpu_80200_name, \
648		cache=xscale_80200_A0_A1_cache_fns
649	xscale_proc_info 80200, 0x69052000, 0xfffffff0, cpu_80200_name
650	xscale_proc_info 80219, 0x69052e20, 0xffffffe0, cpu_80219_name
651	xscale_proc_info 8032x, 0x69052420, 0xfffff7e0, cpu_8032x_name
652	xscale_proc_info 8033x, 0x69054010, 0xfffffd30, cpu_8033x_name
653	xscale_proc_info pxa250, 0x69052100, 0xfffff7f0, cpu_pxa250_name
654	xscale_proc_info pxa210, 0x69052120, 0xfffff3f0, cpu_pxa210_name
655	xscale_proc_info ixp2400, 0x69054190, 0xfffffff0, cpu_ixp2400_name
656	xscale_proc_info ixp2800, 0x690541a0, 0xfffffff0, cpu_ixp2800_name
657	xscale_proc_info ixp42x, 0x690541c0, 0xffffffc0, cpu_ixp42x_name
658	xscale_proc_info ixp43x, 0x69054040, 0xfffffff0, cpu_ixp43x_name
659	xscale_proc_info ixp46x, 0x69054200, 0xffffff00, cpu_ixp46x_name
660	xscale_proc_info pxa255, 0x69052d00, 0xfffffff0, cpu_pxa255_name
661	xscale_proc_info pxa270, 0x69054110, 0xfffffff0, cpu_pxa270_name
v3.15
  1/*
  2 *  linux/arch/arm/mm/proc-xscale.S
  3 *
  4 *  Author:	Nicolas Pitre
  5 *  Created:	November 2000
  6 *  Copyright:	(C) 2000, 2001 MontaVista Software Inc.
  7 *
  8 * This program is free software; you can redistribute it and/or modify
  9 * it under the terms of the GNU General Public License version 2 as
 10 * published by the Free Software Foundation.
 11 *
 12 * MMU functions for the Intel XScale CPUs
 13 *
 14 * 2001 Aug 21:
 15 *	some contributions by Brett Gaines <brett.w.gaines@intel.com>
 16 *	Copyright 2001 by Intel Corp.
 17 *
 18 * 2001 Sep 08:
 19 *	Completely revisited, many important fixes
 20 *	Nicolas Pitre <nico@fluxnic.net>
 21 */
 22
 23#include <linux/linkage.h>
 24#include <linux/init.h>
 25#include <asm/assembler.h>
 26#include <asm/hwcap.h>
 27#include <asm/pgtable.h>
 28#include <asm/pgtable-hwdef.h>
 29#include <asm/page.h>
 30#include <asm/ptrace.h>
 31#include "proc-macros.S"
 32
 33/*
 34 * This is the maximum size of an area which will be flushed.  If the area
 35 * is larger than this, then we flush the whole cache
 36 */
 37#define MAX_AREA_SIZE	32768
 38
 39/*
 40 * the cache line size of the I and D cache
 41 */
 42#define CACHELINESIZE	32
 43
 44/*
 45 * the size of the data cache
 46 */
 47#define CACHESIZE	32768
 48
 49/*
 50 * Virtual address used to allocate the cache when flushed
 51 *
 52 * This must be an address range which is _never_ used.  It should
 53 * apparently have a mapping in the corresponding page table for
 54 * compatibility with future CPUs that _could_ require it.  For instance we
 55 * don't care.
 56 *
 57 * This must be aligned on a 2*CACHESIZE boundary.  The code selects one of
 58 * the 2 areas in alternance each time the clean_d_cache macro is used.
 59 * Without this the XScale core exhibits cache eviction problems and no one
 60 * knows why.
 61 *
 62 * Reminder: the vector table is located at 0xffff0000-0xffff0fff.
 63 */
 64#define CLEAN_ADDR	0xfffe0000
 65
 66/*
 67 * This macro is used to wait for a CP15 write and is needed
 68 * when we have to ensure that the last operation to the co-pro
 69 * was completed before continuing with operation.
 70 */
 71	.macro	cpwait, rd
 72	mrc	p15, 0, \rd, c2, c0, 0		@ arbitrary read of cp15
 73	mov	\rd, \rd			@ wait for completion
 74	sub 	pc, pc, #4			@ flush instruction pipeline
 75	.endm
 76
 77	.macro	cpwait_ret, lr, rd
 78	mrc	p15, 0, \rd, c2, c0, 0		@ arbitrary read of cp15
 79	sub	pc, \lr, \rd, LSR #32		@ wait for completion and
 80						@ flush instruction pipeline
 81	.endm
 82
 83/*
 84 * This macro cleans the entire dcache using line allocate.
 85 * The main loop has been unrolled to reduce loop overhead.
 86 * rd and rs are two scratch registers.
 87 */
 88	.macro  clean_d_cache, rd, rs
 89	ldr	\rs, =clean_addr
 90	ldr	\rd, [\rs]
 91	eor	\rd, \rd, #CACHESIZE
 92	str	\rd, [\rs]
 93	add	\rs, \rd, #CACHESIZE
 941:	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
 95	add	\rd, \rd, #CACHELINESIZE
 96	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
 97	add	\rd, \rd, #CACHELINESIZE
 98	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
 99	add	\rd, \rd, #CACHELINESIZE
100	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
101	add	\rd, \rd, #CACHELINESIZE
102	teq	\rd, \rs
103	bne	1b
104	.endm
105
106	.data
 
107clean_addr:	.word	CLEAN_ADDR
108
109	.text
110
111/*
112 * cpu_xscale_proc_init()
113 *
114 * Nothing too exciting at the moment
115 */
116ENTRY(cpu_xscale_proc_init)
117	@ enable write buffer coalescing. Some bootloader disable it
118	mrc	p15, 0, r1, c1, c0, 1
119	bic	r1, r1, #1
120	mcr	p15, 0, r1, c1, c0, 1
121	mov	pc, lr
122
123/*
124 * cpu_xscale_proc_fin()
125 */
126ENTRY(cpu_xscale_proc_fin)
127	mrc	p15, 0, r0, c1, c0, 0		@ ctrl register
128	bic	r0, r0, #0x1800			@ ...IZ...........
129	bic	r0, r0, #0x0006			@ .............CA.
130	mcr	p15, 0, r0, c1, c0, 0		@ disable caches
131	mov	pc, lr
132
133/*
134 * cpu_xscale_reset(loc)
135 *
136 * Perform a soft reset of the system.  Put the CPU into the
137 * same state as it would be if it had been reset, and branch
138 * to what would be the reset vector.
139 *
140 * loc: location to jump to for soft reset
141 *
142 * Beware PXA270 erratum E7.
143 */
144	.align	5
145	.pushsection	.idmap.text, "ax"
146ENTRY(cpu_xscale_reset)
147	mov	r1, #PSR_F_BIT|PSR_I_BIT|SVC_MODE
148	msr	cpsr_c, r1			@ reset CPSR
149	mcr	p15, 0, r1, c10, c4, 1		@ unlock I-TLB
150	mcr	p15, 0, r1, c8, c5, 0		@ invalidate I-TLB
151	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
152	bic	r1, r1, #0x0086			@ ........B....CA.
153	bic	r1, r1, #0x3900			@ ..VIZ..S........
154	sub	pc, pc, #4			@ flush pipeline
155	@ *** cache line aligned ***
156	mcr	p15, 0, r1, c1, c0, 0		@ ctrl register
157	bic	r1, r1, #0x0001			@ ...............M
158	mcr	p15, 0, ip, c7, c7, 0		@ invalidate I,D caches & BTB
159	mcr	p15, 0, r1, c1, c0, 0		@ ctrl register
160	@ CAUTION: MMU turned off from this point. We count on the pipeline
161	@ already containing those two last instructions to survive.
162	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
163	mov	pc, r0
164ENDPROC(cpu_xscale_reset)
165	.popsection
166
167/*
168 * cpu_xscale_do_idle()
169 *
170 * Cause the processor to idle
171 *
172 * For now we do nothing but go to idle mode for every case
173 *
174 * XScale supports clock switching, but using idle mode support
175 * allows external hardware to react to system state changes.
176 */
177	.align	5
178
179ENTRY(cpu_xscale_do_idle)
180	mov	r0, #1
181	mcr	p14, 0, r0, c7, c0, 0		@ Go to IDLE
182	mov	pc, lr
183
184/* ================================= CACHE ================================ */
185
186/*
187 *	flush_icache_all()
188 *
189 *	Unconditionally clean and invalidate the entire icache.
190 */
191ENTRY(xscale_flush_icache_all)
192	mov	r0, #0
193	mcr	p15, 0, r0, c7, c5, 0		@ invalidate I cache
194	mov	pc, lr
195ENDPROC(xscale_flush_icache_all)
196
197/*
198 *	flush_user_cache_all()
199 *
200 *	Invalidate all cache entries in a particular address
201 *	space.
202 */
203ENTRY(xscale_flush_user_cache_all)
204	/* FALLTHROUGH */
205
206/*
207 *	flush_kern_cache_all()
208 *
209 *	Clean and invalidate the entire cache.
210 */
211ENTRY(xscale_flush_kern_cache_all)
212	mov	r2, #VM_EXEC
213	mov	ip, #0
214__flush_whole_cache:
215	clean_d_cache r0, r1
216	tst	r2, #VM_EXEC
217	mcrne	p15, 0, ip, c7, c5, 0		@ Invalidate I cache & BTB
218	mcrne	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
219	mov	pc, lr
220
221/*
222 *	flush_user_cache_range(start, end, vm_flags)
223 *
224 *	Invalidate a range of cache entries in the specified
225 *	address space.
226 *
227 *	- start - start address (may not be aligned)
228 *	- end	- end address (exclusive, may not be aligned)
229 *	- vma	- vma_area_struct describing address space
230 */
231	.align	5
232ENTRY(xscale_flush_user_cache_range)
233	mov	ip, #0
234	sub	r3, r1, r0			@ calculate total size
235	cmp	r3, #MAX_AREA_SIZE
236	bhs	__flush_whole_cache
237
2381:	tst	r2, #VM_EXEC
239	mcrne	p15, 0, r0, c7, c5, 1		@ Invalidate I cache line
240	mcr	p15, 0, r0, c7, c10, 1		@ Clean D cache line
241	mcr	p15, 0, r0, c7, c6, 1		@ Invalidate D cache line
242	add	r0, r0, #CACHELINESIZE
243	cmp	r0, r1
244	blo	1b
245	tst	r2, #VM_EXEC
246	mcrne	p15, 0, ip, c7, c5, 6		@ Invalidate BTB
247	mcrne	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
248	mov	pc, lr
249
250/*
251 *	coherent_kern_range(start, end)
252 *
253 *	Ensure coherency between the Icache and the Dcache in the
254 *	region described by start.  If you have non-snooping
255 *	Harvard caches, you need to implement this function.
256 *
257 *	- start  - virtual start address
258 *	- end	 - virtual end address
259 *
260 *	Note: single I-cache line invalidation isn't used here since
261 *	it also trashes the mini I-cache used by JTAG debuggers.
262 */
263ENTRY(xscale_coherent_kern_range)
264	bic	r0, r0, #CACHELINESIZE - 1
2651:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
266	add	r0, r0, #CACHELINESIZE
267	cmp	r0, r1
268	blo	1b
269	mov	r0, #0
270	mcr	p15, 0, r0, c7, c5, 0		@ Invalidate I cache & BTB
271	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
272	mov	pc, lr
273
274/*
275 *	coherent_user_range(start, end)
276 *
277 *	Ensure coherency between the Icache and the Dcache in the
278 *	region described by start.  If you have non-snooping
279 *	Harvard caches, you need to implement this function.
280 *
281 *	- start  - virtual start address
282 *	- end	 - virtual end address
283 */
284ENTRY(xscale_coherent_user_range)
285	bic	r0, r0, #CACHELINESIZE - 1
2861:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
287	mcr	p15, 0, r0, c7, c5, 1		@ Invalidate I cache entry
288	add	r0, r0, #CACHELINESIZE
289	cmp	r0, r1
290	blo	1b
291	mov	r0, #0
292	mcr	p15, 0, r0, c7, c5, 6		@ Invalidate BTB
293	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
294	mov	pc, lr
295
296/*
297 *	flush_kern_dcache_area(void *addr, size_t size)
298 *
299 *	Ensure no D cache aliasing occurs, either with itself or
300 *	the I cache
301 *
302 *	- addr	- kernel address
303 *	- size	- region size
304 */
305ENTRY(xscale_flush_kern_dcache_area)
306	add	r1, r0, r1
3071:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
308	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
309	add	r0, r0, #CACHELINESIZE
310	cmp	r0, r1
311	blo	1b
312	mov	r0, #0
313	mcr	p15, 0, r0, c7, c5, 0		@ Invalidate I cache & BTB
314	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
315	mov	pc, lr
316
317/*
318 *	dma_inv_range(start, end)
319 *
320 *	Invalidate (discard) the specified virtual address range.
321 *	May not write back any entries.  If 'start' or 'end'
322 *	are not cache line aligned, those lines must be written
323 *	back.
324 *
325 *	- start  - virtual start address
326 *	- end	 - virtual end address
327 */
328xscale_dma_inv_range:
329	tst	r0, #CACHELINESIZE - 1
330	bic	r0, r0, #CACHELINESIZE - 1
331	mcrne	p15, 0, r0, c7, c10, 1		@ clean D entry
332	tst	r1, #CACHELINESIZE - 1
333	mcrne	p15, 0, r1, c7, c10, 1		@ clean D entry
3341:	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
335	add	r0, r0, #CACHELINESIZE
336	cmp	r0, r1
337	blo	1b
338	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
339	mov	pc, lr
340
341/*
342 *	dma_clean_range(start, end)
343 *
344 *	Clean the specified virtual address range.
345 *
346 *	- start  - virtual start address
347 *	- end	 - virtual end address
348 */
349xscale_dma_clean_range:
350	bic	r0, r0, #CACHELINESIZE - 1
3511:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
352	add	r0, r0, #CACHELINESIZE
353	cmp	r0, r1
354	blo	1b
355	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
356	mov	pc, lr
357
358/*
359 *	dma_flush_range(start, end)
360 *
361 *	Clean and invalidate the specified virtual address range.
362 *
363 *	- start  - virtual start address
364 *	- end	 - virtual end address
365 */
366ENTRY(xscale_dma_flush_range)
367	bic	r0, r0, #CACHELINESIZE - 1
3681:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
369	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
370	add	r0, r0, #CACHELINESIZE
371	cmp	r0, r1
372	blo	1b
373	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
374	mov	pc, lr
375
376/*
377 *	dma_map_area(start, size, dir)
378 *	- start	- kernel virtual start address
379 *	- size	- size of region
380 *	- dir	- DMA direction
381 */
382ENTRY(xscale_dma_map_area)
383	add	r1, r1, r0
384	cmp	r2, #DMA_TO_DEVICE
385	beq	xscale_dma_clean_range
386	bcs	xscale_dma_inv_range
387	b	xscale_dma_flush_range
388ENDPROC(xscale_dma_map_area)
389
390/*
391 *	dma_map_area(start, size, dir)
392 *	- start	- kernel virtual start address
393 *	- size	- size of region
394 *	- dir	- DMA direction
395 */
396ENTRY(xscale_80200_A0_A1_dma_map_area)
397	add	r1, r1, r0
398	teq	r2, #DMA_TO_DEVICE
399	beq	xscale_dma_clean_range
400	b	xscale_dma_flush_range
401ENDPROC(xscale_80200_A0_A1_dma_map_area)
402
403/*
404 *	dma_unmap_area(start, size, dir)
405 *	- start	- kernel virtual start address
406 *	- size	- size of region
407 *	- dir	- DMA direction
408 */
409ENTRY(xscale_dma_unmap_area)
410	mov	pc, lr
411ENDPROC(xscale_dma_unmap_area)
412
413	.globl	xscale_flush_kern_cache_louis
414	.equ	xscale_flush_kern_cache_louis, xscale_flush_kern_cache_all
415
416	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
417	define_cache_functions xscale
418
419/*
420 * On stepping A0/A1 of the 80200, invalidating D-cache by line doesn't
421 * clear the dirty bits, which means that if we invalidate a dirty line,
422 * the dirty data can still be written back to external memory later on.
423 *
424 * The recommended workaround is to always do a clean D-cache line before
425 * doing an invalidate D-cache line, so on the affected processors,
426 * dma_inv_range() is implemented as dma_flush_range().
427 *
428 * See erratum #25 of "Intel 80200 Processor Specification Update",
429 * revision January 22, 2003, available at:
430 *     http://www.intel.com/design/iio/specupdt/273415.htm
431 */
432.macro a0_alias basename
433	.globl xscale_80200_A0_A1_\basename
434	.type xscale_80200_A0_A1_\basename , %function
435	.equ xscale_80200_A0_A1_\basename , xscale_\basename
436.endm
437
438/*
439 * Most of the cache functions are unchanged for these processor revisions.
440 * Export suitable alias symbols for the unchanged functions:
441 */
442	a0_alias flush_icache_all
443	a0_alias flush_user_cache_all
444	a0_alias flush_kern_cache_all
445	a0_alias flush_kern_cache_louis
446	a0_alias flush_user_cache_range
447	a0_alias coherent_kern_range
448	a0_alias coherent_user_range
449	a0_alias flush_kern_dcache_area
450	a0_alias dma_flush_range
451	a0_alias dma_unmap_area
452
453	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
454	define_cache_functions xscale_80200_A0_A1
455
456ENTRY(cpu_xscale_dcache_clean_area)
4571:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
458	add	r0, r0, #CACHELINESIZE
459	subs	r1, r1, #CACHELINESIZE
460	bhi	1b
461	mov	pc, lr
462
463/* =============================== PageTable ============================== */
464
465/*
466 * cpu_xscale_switch_mm(pgd)
467 *
468 * Set the translation base pointer to be as described by pgd.
469 *
470 * pgd: new page tables
471 */
472	.align	5
473ENTRY(cpu_xscale_switch_mm)
474	clean_d_cache r1, r2
475	mcr	p15, 0, ip, c7, c5, 0		@ Invalidate I cache & BTB
476	mcr	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
477	mcr	p15, 0, r0, c2, c0, 0		@ load page table pointer
478	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
479	cpwait_ret lr, ip
480
481/*
482 * cpu_xscale_set_pte_ext(ptep, pte, ext)
483 *
484 * Set a PTE and flush it out
485 *
486 * Errata 40: must set memory to write-through for user read-only pages.
487 */
488cpu_xscale_mt_table:
489	.long	0x00						@ L_PTE_MT_UNCACHED
490	.long	PTE_BUFFERABLE					@ L_PTE_MT_BUFFERABLE
491	.long	PTE_CACHEABLE					@ L_PTE_MT_WRITETHROUGH
492	.long	PTE_CACHEABLE | PTE_BUFFERABLE			@ L_PTE_MT_WRITEBACK
493	.long	PTE_EXT_TEX(1) | PTE_BUFFERABLE			@ L_PTE_MT_DEV_SHARED
494	.long	0x00						@ unused
495	.long	PTE_EXT_TEX(1) | PTE_CACHEABLE			@ L_PTE_MT_MINICACHE
496	.long	PTE_EXT_TEX(1) | PTE_CACHEABLE | PTE_BUFFERABLE	@ L_PTE_MT_WRITEALLOC
497	.long	0x00						@ unused
498	.long	PTE_BUFFERABLE					@ L_PTE_MT_DEV_WC
499	.long	0x00						@ unused
500	.long	PTE_CACHEABLE | PTE_BUFFERABLE			@ L_PTE_MT_DEV_CACHED
501	.long	0x00						@ L_PTE_MT_DEV_NONSHARED
502	.long	0x00						@ unused
503	.long	0x00						@ unused
504	.long	0x00						@ unused
505
506	.align	5
507ENTRY(cpu_xscale_set_pte_ext)
508	xscale_set_pte_ext_prologue
509
510	@
511	@ Erratum 40: must set memory to write-through for user read-only pages
512	@
513	and	ip, r1, #(L_PTE_MT_MASK | L_PTE_USER | L_PTE_RDONLY) & ~(4 << 2)
514	teq	ip, #L_PTE_MT_WRITEBACK | L_PTE_USER | L_PTE_RDONLY
515
516	moveq	r1, #L_PTE_MT_WRITETHROUGH
517	and	r1, r1, #L_PTE_MT_MASK
518	adr	ip, cpu_xscale_mt_table
519	ldr	ip, [ip, r1]
520	bic	r2, r2, #0x0c
521	orr	r2, r2, ip
522
523	xscale_set_pte_ext_epilogue
524	mov	pc, lr
525
526	.ltorg
527	.align
528
529.globl	cpu_xscale_suspend_size
530.equ	cpu_xscale_suspend_size, 4 * 6
531#ifdef CONFIG_ARM_CPU_SUSPEND
532ENTRY(cpu_xscale_do_suspend)
533	stmfd	sp!, {r4 - r9, lr}
534	mrc	p14, 0, r4, c6, c0, 0	@ clock configuration, for turbo mode
535	mrc	p15, 0, r5, c15, c1, 0	@ CP access reg
536	mrc	p15, 0, r6, c13, c0, 0	@ PID
537	mrc	p15, 0, r7, c3, c0, 0	@ domain ID
538	mrc	p15, 0, r8, c1, c1, 0	@ auxiliary control reg
539	mrc	p15, 0, r9, c1, c0, 0	@ control reg
540	bic	r4, r4, #2		@ clear frequency change bit
541	stmia	r0, {r4 - r9}		@ store cp regs
542	ldmfd	sp!, {r4 - r9, pc}
543ENDPROC(cpu_xscale_do_suspend)
544
545ENTRY(cpu_xscale_do_resume)
546	ldmia	r0, {r4 - r9}		@ load cp regs
547	mov	ip, #0
548	mcr	p15, 0, ip, c8, c7, 0	@ invalidate I & D TLBs
549	mcr	p15, 0, ip, c7, c7, 0	@ invalidate I & D caches, BTB
550	mcr	p14, 0, r4, c6, c0, 0	@ clock configuration, turbo mode.
551	mcr	p15, 0, r5, c15, c1, 0	@ CP access reg
552	mcr	p15, 0, r6, c13, c0, 0	@ PID
553	mcr	p15, 0, r7, c3, c0, 0	@ domain ID
554	mcr	p15, 0, r1, c2, c0, 0	@ translation table base addr
555	mcr	p15, 0, r8, c1, c1, 0	@ auxiliary control reg
556	mov	r0, r9			@ control register
557	b	cpu_resume_mmu
558ENDPROC(cpu_xscale_do_resume)
559#endif
560
561	.type	__xscale_setup, #function
562__xscale_setup:
563	mcr	p15, 0, ip, c7, c7, 0		@ invalidate I, D caches & BTB
564	mcr	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
565	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I, D TLBs
566	mov	r0, #1 << 6			@ cp6 for IOP3xx and Bulverde
567	orr	r0, r0, #1 << 13		@ Its undefined whether this
568	mcr	p15, 0, r0, c15, c1, 0		@ affects USR or SVC modes
569
570	adr	r5, xscale_crval
571	ldmia	r5, {r5, r6}
572	mrc	p15, 0, r0, c1, c0, 0		@ get control register
573	bic	r0, r0, r5
574	orr	r0, r0, r6
575	mov	pc, lr
576	.size	__xscale_setup, . - __xscale_setup
577
578	/*
579	 *  R
580	 * .RVI ZFRS BLDP WCAM
581	 * ..11 1.01 .... .101
582	 * 
583	 */
584	.type	xscale_crval, #object
585xscale_crval:
586	crval	clear=0x00003b07, mmuset=0x00003905, ucset=0x00001900
587
588	__INITDATA
589
590	@ define struct processor (see <asm/proc-fns.h> and proc-macros.S)
591	define_processor_functions xscale, dabort=v5t_early_abort, pabort=legacy_pabort, suspend=1
592
593	.section ".rodata"
594
595	string	cpu_arch_name, "armv5te"
596	string	cpu_elf_name, "v5"
597
598	string	cpu_80200_A0_A1_name, "XScale-80200 A0/A1"
599	string	cpu_80200_name, "XScale-80200"
600	string	cpu_80219_name, "XScale-80219"
601	string	cpu_8032x_name, "XScale-IOP8032x Family"
602	string	cpu_8033x_name, "XScale-IOP8033x Family"
603	string	cpu_pxa250_name, "XScale-PXA250"
604	string	cpu_pxa210_name, "XScale-PXA210"
605	string	cpu_ixp42x_name, "XScale-IXP42x Family"
606	string	cpu_ixp43x_name, "XScale-IXP43x Family"
607	string	cpu_ixp46x_name, "XScale-IXP46x Family"
608	string	cpu_ixp2400_name, "XScale-IXP2400"
609	string	cpu_ixp2800_name, "XScale-IXP2800"
610	string	cpu_pxa255_name, "XScale-PXA255"
611	string	cpu_pxa270_name, "XScale-PXA270"
612
613	.align
614
615	.section ".proc.info.init", #alloc, #execinstr
616
617.macro xscale_proc_info name:req, cpu_val:req, cpu_mask:req, cpu_name:req, cache
618	.type	__\name\()_proc_info,#object
619__\name\()_proc_info:
620	.long	\cpu_val
621	.long	\cpu_mask
622	.long	PMD_TYPE_SECT | \
623		PMD_SECT_BUFFERABLE | \
624		PMD_SECT_CACHEABLE | \
625		PMD_SECT_AP_WRITE | \
626		PMD_SECT_AP_READ
627	.long	PMD_TYPE_SECT | \
628		PMD_SECT_AP_WRITE | \
629		PMD_SECT_AP_READ
630	b	__xscale_setup
631	.long	cpu_arch_name
632	.long	cpu_elf_name
633	.long	HWCAP_SWP|HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT|HWCAP_EDSP
634	.long	\cpu_name
635	.long	xscale_processor_functions
636	.long	v4wbi_tlb_fns
637	.long	xscale_mc_user_fns
638	.ifb \cache
639		.long	xscale_cache_fns
640	.else
641		.long	\cache
642	.endif
643	.size	__\name\()_proc_info, . - __\name\()_proc_info
644.endm
645
646	xscale_proc_info 80200_A0_A1, 0x69052000, 0xfffffffe, cpu_80200_name, \
647		cache=xscale_80200_A0_A1_cache_fns
648	xscale_proc_info 80200, 0x69052000, 0xfffffff0, cpu_80200_name
649	xscale_proc_info 80219, 0x69052e20, 0xffffffe0, cpu_80219_name
650	xscale_proc_info 8032x, 0x69052420, 0xfffff7e0, cpu_8032x_name
651	xscale_proc_info 8033x, 0x69054010, 0xfffffd30, cpu_8033x_name
652	xscale_proc_info pxa250, 0x69052100, 0xfffff7f0, cpu_pxa250_name
653	xscale_proc_info pxa210, 0x69052120, 0xfffff3f0, cpu_pxa210_name
654	xscale_proc_info ixp2400, 0x69054190, 0xfffffff0, cpu_ixp2400_name
655	xscale_proc_info ixp2800, 0x690541a0, 0xfffffff0, cpu_ixp2800_name
656	xscale_proc_info ixp42x, 0x690541c0, 0xffffffc0, cpu_ixp42x_name
657	xscale_proc_info ixp43x, 0x69054040, 0xfffffff0, cpu_ixp43x_name
658	xscale_proc_info ixp46x, 0x69054200, 0xffffff00, cpu_ixp46x_name
659	xscale_proc_info pxa255, 0x69052d00, 0xfffffff0, cpu_pxa255_name
660	xscale_proc_info pxa270, 0x69054110, 0xfffffff0, cpu_pxa270_name